Victoria 2022: Lower House Results, Poll Performance And Pendulum Tilt

 LABOR 56 (+1) COALITION 28* (+1*) GREENS 4 (+1) IND 0 (-3)

(Changes are compared to last election)

(* Assuming Narracan is retained)
Current 2PP excluding Narracan 55.00 to ALP 

Projected 2PP treating Narracan as uniform swing 54.83 to ALP
Current 2PP swing accounting for 2PP-uncontested seats 2.70% to Coalition

Following the fast release of 2PP results for every seat it's time to do my usual wrapup of the Victorian lower house election.  The election isn't actually over yet, because of the supplementary election in Narracan to be held early in the new year (and in case there are any further minor changes to the figures), but I think it's best to put it out now with the obligatory cautions.  Throughout this article any use of an asterisk (*) means "subject to Narracan".

This election has presented a salutory lesson for media pundits who like to hype up the chances of hung parliaments, except that overwhelming evidence shows that this stuff will never stop.  It is the fifteenth consecutive Australian state or federal election to return a majority government, and there was serious hung parliament speculation in twelve of those.  Someday (who knows, maybe even NSW?) hung parliament club will be right again, but so will a stopped clock.  In this case despite every single published poll pointing to a Labor majority as being more likely than not (in some cases clearly, in others precariously) many (not all) journalists chose to play up the chance of a hung parliament.   

In this the media were aided by Labor party insiders (a bunch of notorious Eeyores) and Liberal insiders whose "polling" was either grossly defective or made up.  But media also besmirched themselves (the Herald-Sun in particular) by running with obvious nonsense such as an Ian Cook internal "exit poll" with a ludicrous sample size and no evidence of credentials.  There were also several pollsters who did their cause no good by issuing projections that in some cases clashed with their publicly released polling (Redbridge) or that ignored basic projection principles like personal votes (Morgan), while Resolve's problem was partly that its numbers were closer than the election in the first place.   No pollster who attempted projections released any public regional breakdown data so it appears that those pollsters who did this either didn't have such data or else the data were wrong.  Newspoll wisely avoided issuing any seat projection.

This was, in truth, a hard election to project.  That Labor won 56 seats is a surprise to those of us who expected Labor to land around 48-50, but at least we are only surprised rather than very wrong (the latter being the fate of those who tried to outguess the polling and jumped in the wrong direction).  That independents won nothing was more or less unprojectable, but Labor also did very well in keeping losses to the Greens to just one seat and in outperforming the 2PP pendulum to still win a few more seats than would have been expected.  The least surprised were those who modelled the state election by assuming it would be like the federal election (with swings against the government falling harmlessly in the lockdown-blighted northern and western outer suburbs where Labor had massive margins) but normally modelling state elections off federal elections is a recipe for disaster.  Perhaps this one was different because of the extent to which the state situation overshadowed the Victorian federal election campaign.  

Anyway the Victorian media needs to collectively reflect on the damage to its credibility caused by the hung parliament hyping at this election.  We would expect the notoriously biased Herald-Sun to be in the tank for the Coalition (as it was in 2018) but the mystery is why other media so often swallowed its narrative that the Premier was unpopular in spite of there being zero public polling evidence that opinion about him was any worse than divided.  It is as if a disease similar to poll-herding infected the commentariat; nobody wants to miss the boat if a hung parliament actually happened, the Herald-Sun was trumpeting massive trouble for Labor, so other media chose to play safe and have ended up just slightly less wrong than the Herald-Sun.  The result of this is that an imperiously re-elected Premier has been given a mandate to treat journalists like bugs for as long as he wants to continue.  This is a shame because there was some terrific investigative journalism this election that has been let down by the way papers mis-framed the overall narrative.

Final* vote share results

Aside from the Narracan issue I am unsure if there is any further potential for minor corrections, but for now Labor is on 37.03%, Coalition 34.37, Greens 11.50, Independents 5.44 and others 11.66.  Others are headed by Family First Victoria 3.05% and Animal Justice 2.51% (both of these contested every seat) with other contributors including Freedom Party Victoria 1.71% Victorian Socialists 1.35% and Labour DLP 1.17%.  

The current 2PP is, by my count, 55.00% excluding Narracan.  (I get 1991168 - 1629464.) For the second election in a row 2PP swing modelling is messed around by a seat not being available for the 2PP (in 2018 the Liberals did not contest Richmond).  The current 2PP swing in 86 seats excluding Richmond and Narracan is 2.70%.  If Labor contests Narracan, the 2PP will almost certainly drop slightly; by uniform swing it would finish at 54.83.

Labor is receiving about a 0.2% 2PP legup from leakage during three-cornered contests, though I cannot give a precise figure because Labor finished fourth in Shepparton and Mildura.  On this basis Labor is getting about 61.8% of all preferences from non-major-party candidates, down from 67.4% in 2018 (but I am not sure if the latter figure is three-cornered-corrected).  Once again the major driver is that the Greens' share of the preference pool has fallen, this time from 48% to about 40%.  

Victoria does not release 2PP breakdowns by party of origin as is done federally.  The best one can do is estimate the 3CP flows (eg of voters who ranked the Greens ahead of both majors, what proportion ranked Labor ahead of the Coalition?)  Even this is inexact because of cases where the Coalition, Labor and Greens were not the top three, where there was no preference distribution or where the preference distribution halted with multiple parties including the Greens not distributed.  So for 16 seats I have no data on the Green split at all, and for several others I have incomplete data.  These cases to some degree cancel out, since the no-data cases are often divisions where the Greens poll badly and have weak preference flows, but the incomplete-data cases include several teal-challenge seats where the Labor-Liberal share of preferences thrown while a teal is in the mix is likely to underestimate the 2PP flow from the Greens overall.  

Overall I estimate that about 81.5% of voters who put the Greens ahead of the major parties preferenced Labor ahead of the Coalition, which is as good as I can do as a Greens to Labor preference flow for doing last-election preferences off polling.  This leaves 48% as the estimated figure from Others to Labor; these compare with 80% and 53% for 2022.  

I will update the more important figures above once Narracan is finished.

Seat results, swings and personal votes

This election was among the most remarkable cases of swing management by an incumbent government.  The graph below shows the relationship between the pre-election margin and the swing to or against Labor in each seat.  (Seats are colour coded by winner).  

I prefer to talk about seats in terms of occupying party rather than notional status, but notional status works better for graphs.  The VEC had 58 notionally Labor seats but these included Liberal-occupied Caulfield, Hastings and Ripon and Independent-vacated Morwell, while its notionally Liberal seats included Labor-occupied Bass and Bayswater.  Although a uniform swing would have cost Labor six notional marginal seats, Labor in fact won Hastings, Glen Waverley, Bass, Bayswater, Ashwood and Pakenham below the swing line and lost only Morwell above it.  The blue and pink zones above show the notionally shifting seats with the Coalition picking up four in the blue zone and losing three in the pink zone for a net gain of one.  Labor also dropped Richmond to the Greens, finishing on 56.

This was mainly achievable because the swing by margin was so uneven in the Labor seats.  In notional Coalition seats there was a more or less flat 2% swing to Coalition and random variation meant Labor picked up a few.  In Labor seats, however, there was on average zero swing in seats on less than 10% and a swing to Coalition of 4.7% in seats on higher margins, with every seat above 12.3% returning a swing to Coalition.

Apart from Morwell (artificial case) and Nepean (star opposing candidate) Labor didn't have big swings against it in any seat below 10%.  Swings to Coalition were wasted in safe Labor seats, often in the outer suburbs.  

I like to review personal vote effects (a theory and a fact, but not apparent in every election!) but in this case the picture is pretty confused.  Labor did spectacularly well in five of its six "double sophomore" seats (where its incumbent unseated a recontesting Coalition MPs at the previous election) despite losing the sixth (Hawthorn), for an average 3.1% swing to Labor (albeit in seats where Labor would have done OK anyway).  In the "single sophomore" seats (recontesting incumbent, vacant in 2018) there was an average 3.0% swing against Labor (more or less the state average).  In the seats where Labor incumbents retired there was a 4.2% swing, but that is actually better than projected off the graph above as many of these were on large margins.  On the Coalition side there were two sophomore seats with an average 4.2% swing to the party, and two vacancies with an average 1.3%, consistent with expectations.

For what it's worth, my conditional probability model thought that personal vote effects and the distribution of seat margins were worth about 1.6 extra seats to Labor compared to the pendulum prediction for the 2PP that has occurred (ie my model projects 52.6 Labor seats after the loss of Richmond).  To what extent those things have contributed to Labor's beating of the pendulum and how much it is down to regional effects is hard to say, but obviously Labor has done something very right in swing management, by demographic accident or by design.

Non-classic results occurred in four seats won by Labor against the Greens, three seats won by the Greens against Labor, one seat won by the Greens against the Liberals, and four seats won by the Coalition against Independents.  The seat of Mulgrave would probably have had a (lopsided) non-classic result between Daniel Andrews and Ian Cook had it been thrown to preferences, but this has not so far occurred.  As with Werribee at the previous election, I hope the VEC will eventually throw Mulgrave to 2CP as it is a matter of fairness to independents who run second that they should know the swing they need to win, as major party candidates do.

There has as usual been disinformation from some sources on the right about Labor winning with only 37% of the primary vote, ignoring the fact that this was more than the Coalition and that voters for other parties also preferred Labor.  Once again, if these people don't like parties winning a majority on such vote shares, they should campaign for proportional representation or shut up.  Surprisingly at this election only a single Labor seat (Hastings) was won from behind on preferences (the only other two cases of a candidate in second winning being Nationals on Liberal preferences in Mildura and Morwell).  

This election wasn't a massive outlier by the standards of federal drag theory - at least the rich version that takes federal government polling into account.  In August the rich federal drag model was estimating 52 +/-9 seats for Labor assuming a federal 2PP of 57-43 to Labor; Labor won 56.  At the time I thought Labor's federal lead would narrow by now but a quick aggregate of current federal polls suggests it actually hasn't.  Labor has been insulated from the usual anti-government whacking for same-party state governments by the popularity of the Albanese government, but even so has done better in seat terms than the model projects.  (The uneven swing pattern may well be alone enough to account for that without having to throw in "but the opposition ...")

Historically though, this election was:

- the first case of a federally dragged state government gaining seat share from the previous election since Richard Court's Liberals in WA 1996

- the first case of a federally dragged state government gaining seat share past its first term since Rupert Hamer's Liberals in Victoria in 1976.

- the first case of a state government that was eight years old or older gaining seat share since Joh Bjelke-Petersen's Nationals in Queensland in 1986.  (That government was a mere 29 years old at this time.)

Labor vs Greens: Big Swings But No Northcote!

In the seven Labor vs Greens seats, the preference shift caused by the Liberals preferencing the Greens was weaker than might have been expected.  The average Liberal to Green 3CP split by seat was only 64.6% compared to about 43% three seats in 2018. This made an average difference of 4.4% per seat (but only around 3% in Brunswick and Northcote where hardly anybody votes Liberal anyway) and did not decide the result of any contest.  There seems to have been an average 2CP swing to the Greens in these seven seats of around 11.3% (the ABC's swing for Pascoe Vale is greatly exaggerated, I had the seat starting at ALP vs Green 12.5%), meaning that the Greens got an extra 6.9% on top of the Liberal preference shift - but in Northcote (where the Greens gained little from the preference shift) the swing excluding the preference shift was negative (about -1.8%).  

What on earth went wrong for the Greens in Northcote?  Well, a part of it seems to have been a comparatively poor overall Greens and good overall Labor performance.  In the upper house the 2CP swing on the primary votes from Labor to Green (ignoring everyone else) was 4.3% in Northern Metropolitan as a whole, 4.3% in Brunswick (would be slightly redistribution-inflated), 5.2% in Richmond and only 1.2% in Northcote.  So that's about 3%, for whatever reason.  Personal votes would be worth a bit too - in 2018 Northcote was occupied by Lidia Thorpe for the Greens and in 2022 Kat Theophanous (Labor) was the incumbent.  But there is more to it than these things; Theophanous had an unusually strong personal vote, or the Greens candidate or campaign went badly (or Thorpe had a very high personal vote, but I doubt that given that she lost.)  Against a backdrop of a strong election, not winning Northcote (by a whisker) will be a big disappointment for the party.

It's also notable that the large effective 2CP swings to the Greens were not accompanied with primary vote swings.  Excluding Richmond (where the Liberals did not contest in 2018, so a +1.2% primary vote swing was a good result) the average swing on Green primaries in Labor vs Green seats was -0.8%, with a 9.5% primary vote hole in Northcote and gains from Labor in other seats being offset by losses to Victorian Socialists and others.  On the other hand, Sam Hibbins didn't have to win Prahran (Green vs Liberal) from third for a third election in a row, topping the primary count with an 8.1% primary vote swing.

No The Nats Did Not Cost The Liberals Bass

The weak preference flow from the Nationals to the Liberals in Bass (68.8%) has attracted some attention, and it even seems some internal polls misread the seat by not polling and distributing the Coalition parties separately.  A couple of percentage points of flow here would be explained by the Nats getting the donkey vote which flowed to Labor, and the flow is also only a 3CP flow (including all the votes the Nats got from other sources.)   So the flow from the Nats candidate himself might have really been more like 75, which is hardly different to many federal cases around the 80 mark.

The Nats always get blamed when this stuff happens but there's no evidence that any serious number of voters who vote National and preference Labor would have put the Liberals ahead of Labor if the National wasn't running.  (There have been some similarly spurious attempts from Labor rusted-ons to blame teals for Liberal wins in Hawthorn and Kew).  It's simply the case that not all Nationals voters like the Liberal Party, and at an election where the Nationals did well while the Liberals went backwards, that shouldn't be surprising.  Candidate factors can also come into play (the Nats candidate is a former Mayor).  The flow in Bass is virtually identical to that in Ripon 2014 where the leading primaries were Labor 35.1 Liberal 32.8 National 18.3 Green 7.1 and the Liberal 2PP at 50.75 ended up lower than the Coalition primary vote!

Tealflop!

The much vaunted teal independent challenge at this election won nothing.  Teal independents finished up with just one single second place (albeit nearly a win, in Mornington) and other publicised challengers didn't even make the top two.  So-called purple independents in western Melbourne were much hyped but didn't much trouble the scorers and two rural independents lost in the first cases of incumbent full-term indies losing in a state or federal election since 2014.

The teal federal success proved not to be transplantable to Victorian state politics, at least this time around.  Constraints on spending have been cited as a factor but I also think the teal movement lacked the relevance it had during the federal campaign and in particular the ability to pry enough protest votes from the Liberals to succeed.  (In some seats the teal to Labor flows were very nearly as strong as the other way round.) The fact that a lot of the seats targeted by teals were seats where Labor was competitive didn't seem to help, but given that the teals were all losing the 2CP anyway this only meant they were finishing third instead of second.  It was symptomatic of the lack of oxygen that one of the biggest sources of teal coverage in the media was them fighting with the VEC about how to vote cards - they won the fight but I doubt it was a votewinner.

During the campaign I had a look at a range of state and federal compulsory preferencing elections to see how predictive the Independent vote was of seat results, but for this election not only was the independent vote lower than expected (though Resolve's 6% estimate was close), but also the result was an outlier even by Victorian standards, so this projection attempt failed.  Here's the graph again with this year added showing the extent of the flop:

(Line of fit has not been moved from pre-election graph.)

Final Poll Performance* (Interim Assessment)

Normally I wait until every vote is in and final before doing one of these but Narracan is testing my patience, especially as it will only affect the official 2PP if Labor actually contests it (and if Labor does contest it it is likely to behave more like a by-election than a normal seat, which will add a random element that pollsters could not have anticipated).  Following the way I handled Richmond in 2018, I will apply uniform 2PP swing for Narracan in the event of Labor not contesting.

Firstly, this was a great election for statewide final polling and anyone who has said otherwise should be heard no further on the subject of poll accuracy.  The purpose of voting intention polling is to predict vote share, not seat share, which is an exercise for modellers.  Some pollsters played modeller and got their fingers burned, but that is not a reflection on their polls.  The final polls by the last three pollsters in field (Roy Morgan, Newspoll and Resolve) on average got the current primary votes for all of the Coalition, Labor, Greens and Others right to within 0.5% and only slightly underestimated the 2PP (by just under 1%, and that will come down if Labor contests Narracan).  Of the two outfits that only polled once a few weeks out, Freshwater Strategy had an impressive public polling debut as it just about nailed the primary votes for both majors, its biggest issue being the breakdown between Greens and Others (which may well have been a matter of genuine shift anyway).  Redbridge (taken a few weeks out) suffers by comparison with the rest and by being well off on the Coalition vote and the Green/Others breakdown (to which the same comment as Freshwater applies), but it wasn't terrible either.  

As in the federal election I use four different indicators of final poll accuracy:

RMSQ: root mean square error on primary votes.  Punishes outliers

RMS2: the same but with the error on the 2PP given a 50% weighting, because I think 2PP is an important part of an Australian pollster's job

AVE: average error on primary votes

AVE2: average error weighted 50% for 2PP and 50% for average error on primary votes (this is my primary indicator)

The following is a provisional accuracy table assuming uniform primary vote and 2PP swing in the division of Narracan.  

I have omitted an apparently wildly inaccurate Lonergan poll as it is unclear whether the voting intention figures (which had the Greens on 19%) were intended to be scaled or a raw sample prior to scaling.  A small Essential sample released in September is also not included (too far out).  

In this table Morgan has topped AVE2, which is the primary indicator that I use for sorting, but it is possible that Newspoll could top AVE2 if the Coalition's primary vote is very high (which it may well be given that it is a three-cornered contest; on the other hand it is also a vacancy).  A final table will be added to this thread when Narracan is finished.

Newspoll has topped two of the other indicators, narrowly, but Morgan could top these if Labor does better than expected.  Resolve tops one of the indicators (RMSQ) though that ignores 2PP accuracy which I consider to be very important. It is also very close to topping the other primary vote indicator, and could yet do so - its main problem is that its 2PP estimate underestimates the flow to Labor.

In any case Morgan and Newspoll are the most accurate final polls for this election, however Morgan has got a little bit lucky here with errors cancelling out (as it did in its 2013 federal 2PP bullseye).  It was too low on the Coalition primary but its preferencing method greatly underestimated the flow of preferences to Labor, and the resulting 2PP reading was excellent.   It should also be noted that if Labor does not contest Narracan, Morgan will have (on current numbers) exactly nailed the official 2PP.  

I have been very critical of Morgan SMS polls in the past as they were very volatile and inaccurate at a few elections in the mid-2010s, and because the concept seems prone to "motivated response", but this one has performed very well indeed and suggests that the method can be viable.  

Also of note:

1. Resolve was the only poll to issue clearly defined breakdowns for Independent and Other.  At 6% and 12% respectively these were very accurate.

2. Morgan issued an odd breakdown of Others primaries with 1% for DHJP (which polled 0.2%), 0.5% for UAP (which didn't run) and 4.5% for "teal independents" (a large overestimate; I get recognised teals on 1.3%).  

3. The penultimate Morgan (57-43) and Resolve (59-41) polls showed massive Labor leads whereas the penultimate Newspoll was 54-46.  Labor coming down off the massive leads fuelled a view that there was a rapid plunge in voting intention, which in turn played into the hung parliament hoopla, but the final Newspoll said otherwise and was correct.  I suspect there was actually not such a big shift and that some of these polls with huge Labor leads were not accurate.

What was really lacking at this election was any kind of regional polling breakdown that could point to the uneven nature of the swing.  Had such existed, it may well have improved the narrative.  There were also no public seat polls during the campaign.  

Tilted Pendulum: Is Victoria The New South Australia?

The result of this election is that the Coalition faces a very difficult pendulum going into the 2026 election.  I will issue my own post-election 2PP pendulum after Narracan (I dislike versions that mix non-classic and classic seats together) but this is a uniform swing markoff of Labor seats that the Coalition might want to be capturing in its quest for victory in 2026:

Assuming there is no net change in the Greens seats, the Coalition needs a uniform 6.92% swing (52.09%) to reach the orange zone and deprive Labor of a majority.  7.55% (52.72%) reaches the grey zone where the Coalition holds half the parliament and 8.04% (53.21%) puts the Coalition into majority.

To make things even harder, Labor has a number of new MPs in seats around this section of the pendulum, which will make those seats a little but harder to budge than their margins suggest.  

Going into this election the Victorian pendulum already favoured Labor.  The median 2PP was 59.42%, meaning that all else being equal on a uniform swing to a 50-50 2PP the median seat 2PP would still be 51.84% to Labor.  Now the pendulum is even more tilted - a uniform swing of 4.83% to get a 2PP of 50-50 leaves the median 2PP at 52.97 to Labor.  So there is a tilt of about three percent in Labor's favour.

It is figures like this that make it clear that Labor won this election by an enormous margin and that a hung parliament was simply never on the cards.  Labor was expected to need approaching 52% 2PP for a majority but could actually have got one off even 48%.  

Thinking of seats as being, for instance, Labor +3 if Labor's 2PP runs 3 points above the state 2PP, what we have seen over recent elections is a shovelling out of seats around the zero mark and also a narrowing in Labor seats on big margins.  The result is that there is now a big pack of seats around Labor +5, while the average Coalition seat is around Coalition +10. If there was a 50-50 2PP, much of the Coalition's 2PP vote would be wasted in safe seats and narrow losses.  This has been brought about by the realignment of safe Labor seats in the outer suburbs towards the median, while higher income/educated seats in south-eastern Melbourne especially move more to the left.  

To whatever extent through demographics and to whatever extent through political skill, Victoria has now developed a pendulum tilted in Labor's favour, similar to that which favoured Labor in South Australia for a while.  South Australia, with its Playmander legacy, introduced a 2PP-based "fairness clause" for redistributions but it was extremely difficult to implement and generally did not work. As it happened the final redistribution under it worked very well, by which stage it had been abolished.  Perhaps if the realignment accelerates, more once safe Labor seats will become genuine marginals, or perhaps this election will prove to be a temporary response to COVID issues and there will be a drift back to normality. But for the moment it is very hard for the Coalition to get back into government without a very large 2PP win or a favourable swing distribution.  If the Coalition can have an election where it gets a big swing in one go (either winning or nearly so) then that might give it enough personal vote bonuses to offset this tilt to a degree, but I am not sure it is going away entirely in a hurry.

This issue hung over the recent Liberal leadership contest where the party had the choice between John Pesutto who has potential to repair the damage in the inner city and Brad Battin (a more strategically radical option to try to keep driving the outer suburban realignment); they chose Pesutto, for now, by one vote.  

I hope over the next few days to finish a wrapup of the upper house.  [Edit: it is up now.]

Update Sunday: The VEC has announced that in February it will conduct full preference distributions for seats won with an absolute majority, which will provide an Andrews vs Cook 2CP in Mulgrave in the event that Cook does succeed in making the 2CP.   However it is not as yet clear whether this will also occur for seats where there is an incomplete distribution because of someone reaching 50% midway through the distribution.  (These are not a big problem for cases where only three candidates remain, because then the preference flow can be inferred from the 2CP, but they are more of an issue for cases with four or more candidates.)

Monday: The VEC has now advised distributions will occur in all districts.

Update for Narracan (16 Feb):

This is the table as above after including the primary votes for Narracan (which Labor did not contest).  Morgan finishes as the most accurate poll including the 2PP.  Resolve had the most accurate primary votes on either indicator but was well down the list once including the very important 2PP.  There is still some possibility of microscopic changes in the 2PP as all distributions are conducted but that would not affect the outcome.